基于空间截面投影的Otsu图像分割算法

doi:10.3969/j.issn.1000-436x.2014.05.010

摘要/Abstract

摘要：

针对传统三维 Otsu 法存在抗噪性差、计算复杂度高、难以多阈值扩展等不足，提出了一种基于空间截面投影的 Otsu 法。该方法采用垂直于主对角线的平面截三维直方图，并将截面内的点投影到主对角线上，进而建立一维的截面投影直方图以降低计算复杂度，然后基于该一维直方图进行 Otsu 分割；对于分割结果，运用基于阈值的后处理策略做进一步的处理以增强算法的抗噪性。实验结果表明，与现有三维 Otsu 法相比，该算法计算效率高、抗噪性好，能对含不同噪声的图像进行较好的分割。

关键词: 图像分割, 阈值选取, 最大类间方差, 三维Otsu法, 后处理策略

Abstract:

Due to the disadvantages existing in the conventional three-dimensional Otsu (3D Otsu) method, such as poor anti-noise capability, high computational complexity and difficulty to extend to multilevel thresholding, a new Otsu thresholding method based on projection of cross section was proposed. A plane perpendicular to the main diagonal is used to cross the 3D histogram. Then the projections to the main diagonal from the points in the cross section, are used to establish a new histogram. Based on the histogram, the image can be segmented by 1D Otsu. In order to further enhance the anti-noise capability, the post-processing strategy based on threshold is applied to the processed image. The experi-mental result shows that this method has significant improvements in computational efficiency and anti-noise capability, and can segment the images corrupted with different noises well.

Key words: image segmentation, threshold selection, maximum between-class variance, 3D Otsu's method, post-proc-essing strategy

范朝冬,欧阳红林,肖乐意. 基于空间截面投影的Otsu图像分割算法[J]. 通信学报, 2014, 35(5): 70-78.

Chao-dong FAN,Hong-lin OUYANG,Le-yi XIAO. Otsu thresholding method based on projection of cross section for image segmentation[J]. Journal on Communications, 2014, 35(5): 70-78.

图/表 10

图1

图2

图3

图4

图5

图6

图7

图8

图9

表1

参考文献 17

[1]	李伟斌，高二，宋松和 . 一种全局最小化的图像分割方法[J]. 电子与信息学报， 2013,35(4):791-796. LI W B , GAO E , SONG S H . A global minimization method for image segmentation[J]. Journal of Electronics ＆ Information Technology, 2013,35(4):791-796.
[2]	OTSU N . A threshold selection method from gray-level histograms[J]. IEEE Transactions on System Man and Cybernetic, 1979,9(1): 62-66.
[3]	岳峰，左旺孟，王宽全 . 基于分解的灰度图像二维阈值选取算法[J]. 自动化学报， 2009,35(7):1022-1027. YUE F , ZUO W M , WANG K Q . Decomposition based two- dimen-sional threshold algori-thm for gray images[J]. Acta Automatica Sinica, 2009,35(7):1022-1027.
[4]	吴一全，张金矿 . 二维直方图θ划分最大Shannon熵图像阈值分割[J]. 物理学报， 2010,59(8):5487-5495. WU Y Q , ZHANG J K . Image thresholding based on θ-division of 2-D histogram and maximum Shannon entropy[J]. Acta Physica Sinica, 2010,59(8):5487-5495.
[5]	HEN Q , ZHAO L , LU J , et al. Modified two-dimensional Otsu image segmentation algorithm and fast realization[J]. IET Image Processing, 2012,6(4): 426-433.
[6]	阿里木·赛买提，杜培军，柳思聪 . 基于人工蜂群优化的二维最大熵图像分割[J]. 计算机工程， 2012,38(9):223-243. ALIM S , DU P J , LIU S C . Maximum 2D entropy image segmentation based on artificial bee colony optimization[J]. Computer Engineering, 2012,38(9):223-243.
[7]	HORNG M H , LIOU R J . Multilevel minimum cross entropy thresh-old selection based on the firefly algorithm[J]. Expert Systems with Applications, 2011,38(12): 14805-14811.
[8]	GAO H , KWONG S , YANG J J , et al. Particle swarm optimization based on intermediate disturbance strategy algorithm and its applica-tion in multi-threshold image segmentation[J]. Information Sciences, 2013,250(20): 82-112.
[9]	刘健庄，栗文清 . 灰度图像的二维Otsu自动阈值分割法[J]. 自动化学报， 1993,19(1):101-105. LIU J Z , LI W Q . Automatic thresholding of gray-level pictures using two-dimension Otsu method[J]. Acta Automa-tica Sinica, 1993,19(1):101-105.
[10]	景晓军，李剑峰，刘郁林 . 一种基于三维最大类间方差的图像分割算法[J]. 电子学报， 2003,31(9):1281-1285. JING X J , LI J F , LIU Y L . Image segmentation based on 3-D maxi-mum between-cluster variance[J]. Acta Electronica Sinica, 2003,31(9):1281-1285.
[11]	范九伦，赵凤，张雪峰 . 三维Otsu阈值分割方法的递推算法[J]. 电子学报， 2007,35(7):1398-1402. FAN J L , ZHAO F , ZHANG X F . Recursive algorithm for three-di-mensional Otsu's thresholding segmentation method[J]. Acta Elec-tronica Sinica, 2007,35(7):1398-1402.
[12]	STHITPATTANAPONGSA P , SRINARK T . An equivalent 3D Otsu's thresholding method[A]. The 5th Pacific Rim Symposium on Image and Video Technology[C]. 2011. 358-369.
[13]	吴一全，吴文怡，潘喆 . 基于二维直方图斜分的最小类内方差阈值分割[J]. 仪器仪表学报， 2008,29(12):2651-2657. WU Y Q , WU W Y , PAN Z . Minimum within-class variance thresh-olding based on two-dimensional histogram oblique segmentation[J]. Chinese Journal of Scientific Instrument, 2008,29(12):2651-2657.
[14]	张新明，李振云，孙印杰 . 快速二维直方图斜分最小误差的图像阈值分割[J]. 电光与控制， 2012,19(6):8-12. ZHANG X M , LI Z Y , SUN Y J . Fast image thresholding with mini-mum error based on 2-D oblique segmentation[J]. Electronics Optics＆ Control, 2012,19(6):8-12.
[15]	The USC-SIPI image database[EB/OL]. , 2014.
[16]	ADILJAN Y , YOSHIHIRO H , TASUKU M , et al. 2-D direction histo-gram based entropic thresholding[J]. Neurocomputing, 2013,120(23): 287-297.
[17]	张弘，范九伦 . 二维 Arimoto 熵直线型阈值分割法[J]. 光子学报， 2013,42(2):234-240. ZHANG H , FAN J L . wo-dimensional Arimoto entropy linear-type threshold segmentation method[J]. Acta Photonica Sinica, 2013,42(2):234-240.

分割算法	含椒盐噪声的Lena			含高斯噪声的Lena			含混合噪声的Lena
分割算法	区域一致性	错分率	时间/s	区域一致性	错分率	时间/s	区域一致性	错分率	时间/s
传统三维Otsu法	0.973 3	0.054 6	203.721 3	0.944 5	0.126 4	203.931 6	0.921 1	0.200 8	203.740 2
快速三维Otsu法	0.973 3	0.054 6	17.790 6	0.944 5	0.126 4	18.038 4	0.921 1	0.200 8	17.968 1
基于分解的三维Otsu法	0.981 1	0.050 9	4.003 8	0.949 3	0.125 5	4.043 1	0.939 7	0.152 3	4.037 4
中值滤波Otsu法	0.991 1	0.017 7	1.861 4	0.952 7	0.104 5	1.868 1	0.947 0	0.121 0	1.869 7
均值滤波Otsu法	0.958 4	0.090 1	1.021 7	0.959 3	0.088 2	1.039 7	0.946 1	0.123 8	1.034 2
基于截面投影的Otsu法	0.984 0	0.031 7	2.770 8	0.960 1	0.086 2	2.788 7	0.953 1	0.104 8	2.782 9